A circuit card is the current state of the art for building assemblies of electronic devices including a plurality of integrated circuits (“chips”). These assemblies can be separated into multiple types: organic multilayer laminated printed wire board (PWB), low temperature co-fired ceramic (LTCC), and high temperature co-fired ceramic (HTCC). Using each of these technologies, circuit card assemblies have been fabricated.
In a superconducting supercomputer, many of the operating processing integrated circuits (“chips”) are cooled to about 4K, but certain of the memory chips instead have a much warmer operating temperature of about 77K. Providing cooling at 4K is a costly activity, so every effort is made in superconducting supercomputer design to reduce the thermal parasitic load. This includes placing the assembly in vacuum (no convection), use of coatings and multilayer insulation to reduce radiation, and limiting the conductive thermal load between the “hot side” and “cool side” of the entire assembly.
For large scale applications, the state of the art currently solves the problem of achieving the desired operating temperatures for a superconducting supercomputer while avoiding thermal parasitic load by using dewars for each of the cryogenic temperatures. A 4K dewar is maintained with liquid helium and a 77K dewar uses liquid nitrogen. Signals between the two temperature sides are completed by cabling. This solution requires cables that are long from a digital perspective, which results in significant latency between the 4K and 77K regions and would require more parts in the 4K stage. These additional parts consume significant power and make certain designs of superconducting supercomputers infeasible.
In small scale applications, a cryocooler can be used for both temperatures. The intermediate stage of the cryocooler provides a 77K platform while the final stage of the cryocooler provides a 4K stage. Connections between the two zones are completed by cabling. While this brings the two temperatures sides closer together, this approach is not scalable to large applications.